Models and solution methods for large-scale industrial mixed integer programming problems

Abstract: This thesis deals with large-scale industrial problems that can be formulated using mixed integer linear programming (MIP) models. Because of the large problem size, it is not often possible to apply standard solution methods. Therefore special techniques must be used. In this thesis, both full optimization and optimization based heuristic approaches are developed.The body of this thesis consists of five research papers. In the papers we consider industrial cases based on three applications: production planning at Södra Cell AB, ship routing and distribution at Södra Cell AB, and staff routing and scheduling in the Swedish home care.We develop two large-scale MIP models for the production-planning problem. For solving, we use both a direct approach, and a combination of column generation and constraint branching. The latter is a technique to control the branching rules in a branch and bound framework and takes into account application specific properties.For the ship routing problem, we present a MIP model and develop two solution methods. The first is based on an iterative rolling time horizon. In each step a part of the model is solved and later fixed. This is repeated until the full problem is solved. The second approach is to combine a genetic algorithm and linear programming. From the MIP model, we obtain solution bounds rarely present in genetic algorithm frameworks.In the staff routing problem there are special synchronization constraints and we introduce a new MIP model. An exact solution approach based on column generation and branch and bound is developed. We also suggest a variable fixing heuristic, which we use for solving the more general problem with requirements on equal workload.In the computational experiments, we use data instances obtained from real-world cases, and generated instances based on real-world cases. The numerical results show that high quality solutions can be obtained within reasonable time limits in all applications.